Background

Apart from the expression of hormone and human epidermal growth factor receptors there are as yet hardly any predictive factors for treatment efficacy in patients with metastatic breast cancer (MBC) despite a rapidly growing number of treatment options. In this situation it is of utmost importance to identify early indicators of response to systemic treatment to avoid unnecessary exposure to ineffective but toxic therapies and to enable prognostication of progression-free survival (PFS) and overall survival (OS). Circulating tumor cells (CTCs) have been detected in 40–60% of patients with MBC using the CellSearch™ system (Veridex) [1, 2]. The presence of CTCs at levels ≥ 5/7.5 ml peripheral blood is associated with decreased PFS and OS [2–4]. It has been suggested that CTCs provide more clinically relevant information than conventional imaging studies regarding therapeutic efficacy and ultimate outcome [5]. In addition, the prognostic information of ≥ 5 CTCs/7.5 ml blood might be helpful in identifying those patients who would likely experience a worse outcome with standard treatment and might benefit from more aggressive therapy [4]. Thus far, several retrospective and a few prospective studies in patients with MBC have demonstrated the usefulness of monitoring therapeutic efficacy by serial CTC enumerations [6–9]. To further address this important issue, the present study aimed to prospectively assess in a large group of patients whether CTC status at baseline (CTCBL) and after one cycle of a new line of treatment (CTC1C) and changes in CTC status from baseline to completion of one treatment cycle (CTC kinetics, CTCKIN) could serve as early predictors of efficacy in terms of response, PFS, and OS.

Methods

Patients and study design

This was a prospective, single-center, non-randomized, partially blinded, treatment-based study. The study was blinded in the following respects. Both patients and treating physicians were blinded to CTC status, and hence treatment regimens did not depend on CTC status. All investigators and technical staff who performed or reviewed the CTC studies were blinded to patient history and treatment. CTC enumeration and characterization were confirmed by independent reviewers. All radiologists performing computed tomography (CT) scans and magnetic resonance imaging (MRI) studies were blinded to the patient’s treatment regimen. The study was conducted at the National Center for Tumor Diseases (NCT), Heidelberg, Germany and the Department of Obstetrics and Gynecology, University of Heidelberg, Heidelberg, Germany.

Patients included in the study were women with MBC about to start a new line of systemic treatment. Patients were enrolled consecutively between March 2010 and December 2013. Main eligibility criteria were clinical and radiological evidence of measurable or evaluable metastatic disease according to the Response Evaluation Criteria in Solid Tumors (RECIST) criteria [10], age > 18 years, progressive metastatic disease, CTC assessment at baseline, and written informed consent. Before starting a new line of systemic treatment, patients underwent CTC enumeration to determine CTCBL status, defined as positive (CTCBL+) for ≥ 5 CTC or negative (CTCBL-) for < 5 CTC per 7.5 ml of peripheral blood [11]. Determination of CTC status was repeated after the first cycle of treatment (CTC1C). After approx. 3 months, patients were evaluated for response by CT and MRI, as appropriate. Response was defined as complete response (CR), partial response (PR), stable disease (SD), and progressive disease (PD) according to the RECIST criteria, version 1.1 [10]. Evaluation was repeated according to the RECIST criteria every 2–3 months until progression of disease. Survival status was recorded until death or loss to follow-up.

All study procedures, including laboratory evaluations, imaging studies, and treatment planning, were carried out at the NCT, Heidelberg, Germany and the Department of Obstetrics and Gynecology of the University of Heidelberg, Heidelberg, Germany in collaboration with the German Cancer Research Center (DKFZ), Heidelberg, Germany, the Department of Tumor Biology, University Medical Center Hamburg-Eppendorf, Hamburg, Germany, and the Heidelberg Institute for Stem Cell Technology and Experimental Medicine (HI-STEM), Heidelberg, Germany. Ethical approval was obtained from the Ethics Committee of the Medical Faculty of the University of Heidelberg.

CTC enumeration

For CTC enumeration, 7.5 ml peripheral whole blood was collected in a standard 10-ml tube containing ethylenediaminetetraacetic acid (EDTA) and a cellular preservative. Blood samples were kept at room temperature for ≤ 72 hours before analysis using the CellSearch™ assay (CellSearch™ Epithelial Cell Kit/CellSpotter™ Analyzer, Veridex LLC, Raritan, NJ, USA). Sample processing and analysis were done strictly according to the manufacturer’s instructions. The assay uses a ferrofluid coated with antibodies to epithelial cell adhesion molecule (EpCAM) to immunomagnetically separate cells of epithelial origin from blood, and fluorescent staining to differentiate between debris, hematopoietic cells, and epithelial-derived circulating tumor cells [12]. It provides high intra-observer, inter-observer and inter-instrument agreement [2, 13]. Thus, CTCs enumerated and characterized in this study were cells with positive nuclear staining expressing cytokeratin (CK) 8, 18, and 19, and lacking CD45 [11, 14]. Assay operators were trained to classify images generated by the CellSpotter™ Analyzer before study initiation. Samples with < 5 CTCs/7.5 ml were classified as CTC-, those with ≥ 5 CTCs/7.5 ml as CTC+ [11]. CTC kinetics (CTCKIN) were defined in terms of changes in CTC status from CTCBL to CTC1C and categorized as favorable (CTCBL- to CTC1C- and CTCBL+ to CTC1C-) or unfavorable (CTCBL- to CTC1C+ and CTCBL+ to CTC1C+).

HER2 status

Human epidermal growth factor receptor 2 (HER2) status was determined using the immunohistochemistry-based HERCEP™ test (DAKO, Glostrup, Denmark) for semi-quantitative detection of HER2 expression in breast cancer tissue. Expression of HER2 was scored on a scale from 0 to 3+. Tissue samples with a score of 3+ were considered HER2-positive. Whenever the score was 2+, HER2 amplification was determined by fluorescence in-situ hybridization using the Pathvysion Kit (Vysis Inc., Downers Grove, IL, USA).

Data analysis and statistics

Patient demographic and clinical characteristics were summarized as medians and ranges or numbers and percentages, as appropriate. The numbers of missing values were given in ‘no data’ categories. Differences between the CTC+ and CTC- groups were compared using the Wilcoxon rank test and Fisher’s exact test, as appropriate. PFS was defined from date of enrollment until the date of disease progression or death from any cause, whichever occurred first. OS was calculated from the date of enrollment until the date of death from any cause. Patients who were alive or showed no progression at last follow-up were regarded as censored observations. Median follow-up time was calculated using the reverse Kaplan-Meier method.

To identify predictors of PFS and OS, the following candidate predictors were selected a priori based on previous studies and univariate analysis: CTCBL status (negative or positive), age at study entry, molecular subtypes (hormone receptor (HR)+/HER2-, HER2+, or triple negative breast cancer (TNBC)), site of metastasis (local, bone/visceral, or both), number of metastatic sites (one or at least two), and line of therapy (first, second, or at least third). The prognostic effects of these factors were determined by multivariate analysis using a Cox proportional hazards regression model. Patients with missing values in these variables were not included in the Cox regression models. Separate models for CTCBL and CTCKIN were formulated because the CTCBL model showed a fairly larger sample size and to avoid multicollinearity (since CTCBL and CTCKIN are related). Concordance indices were used to estimate the predictive accuracy of the Cox models.

During the initial phase of the study, which comprised the first 100 patients, CTC1C status was routinely determined only in CTCBL+ patients and not in CTCBL- patients. However, as preliminary CTC1C results from CTCBL- patients also drew interest, it was decided to determine CTC1C status in all subsequent patients. This change may have introduced a potential source of bias in the CTC1C results, e.g. proportions. All CTCKIN findings were, thus, conditioned on survival up to the determination of CTC1C status.

Statistical analyses were performed using R (version 3.0.0, package survival). All reported P values were two-sided and a significance level of 5% was chosen.

Results

Patients and study design

From March 2010 through December 2013, 403 consecutive patients were enrolled in the study. Figure 1 shows the flow of patients through the study. Reasons for exclusion from, or non-availability for, further analysis are detailed in the figure legend. Of the 393 evaluable patients with CTCBL counts, 133 (34%) were CTCBL+ and 260 (66%) were CTCBL-. The two patient groups did not differ significantly in median age (range) at initial diagnosis of breast cancer (50 (28–81) vs. 51 (23–79) years) but age at study entry was significantly lower in CTCBL+ patients (57 (33–81) vs. 61 (29–89) years). Patient characteristics at baseline and after one cycle of treatment are summarized in Table 1. Notably, the majority of patients had ER+ (271/378 (72%)), PgR+ (240/370 (65%)), and HER2- (274/341 (80%)) primary tumors. Most patients had more than one metastatic site (305/393 (78%)) and approximately half of patients had both bone and visceral/local metastases (191/393 (49%)). At study entry, 135/391 (35%) patients were about to start third- or higher-line treatment.

Figure 1
figure 1

Flow of patients through the study. Of 403 consecutive patients assessed for eligibility, 10 (2.5%) were excluded from the study because essential data items were not available (no clinical data: 1 patient; no CTCBL data: 9 patients). Of the 393 patients included in the study, 192 had no CTC1C counts and were therefore excluded from further analysis for the following reasons. During the initial phase of the study, i.e. the first 100 patients, CTC1C status was routinely determined only in CTCBL+ patients, resulting in 64 CTC1C- patients without CTC1C counts. Of the remaining 128 patients without CTC1C counts, 12 were excluded because blood samples were not obtained within the predefined study timeframe of 0.5–3.2 months, 25 did not survive to CTC1C assessment because they died within the first 3.2 months, and 91 patients who survived beyond 3.2 months after inclusion had no CTC1C count (41 had not yet proceeded to CTC1C and 50 were lost to follow-up blood sampling as our center often treats external patients).

Full size image

CTC status and response

CTC1C status was assessed after a median (range) of 1.2 (0.5–3.2) months. CTC1C status was positive in 57/201 (28%) and negative in 144/201 (72%) of patients. During the initial phase of the study, which comprised the first 100 patients, CTC1C status was determined only in CTCBL+ patients. As shown in Table 1, at least SD (i.e. CR, PR, or SD) was seen in 162/255 (64%) patients at the 3-month radiological examination, of whom 52/162 (32%) were CTCBL+ while 110/162 (68%) were CTCBL-. Radiological restaging was performed a median of 2.9 (0.5–4.8) months after study entry. PD occurred in 93/255 (36%) patients, of whom 40/93 (43%) were CTCBL+ while 53/93 (57%) were CTCBL- (Fisher’ exact test, P = 0.104). CTCKIN could be determined in 201 patients as both their CTCBL and CTC1C data were available. At least SD was achieved in 55/75 (73%) patients with CTCKIN from CTCBL- to CTC1C-, 21/32 (66%) with CTCKIN from CTCBL+ to CTC1C-, 20/41 (49%) with CTCBL+ to CTC1C+, and 3/6 (50%) with CTCBL- to CTC1C+ (Fisher’s exact test, P = 0.04997).

Table 1 Patient characteristics by CTC+ status at baseline (BL) and after one cycle of treatment (1C)
Full size table

CTC status and survival

Follow-up data were available for 356 patients with a median [95% CI] follow-up of 26.0 [23.7–28.5] months for OS.

Figure 2 shows Kaplan-Meier plots for PFS and OS by CTC status at baseline (CTCBL, top panels) and after the first cycle of a new line of systemic therapy (CTC1C, bottom panels). Median [95% CI] PFS and OS were significantly shorter in CTCBL+ than in CTCBL- patients (PFS: 4.7 [3.7–6.1] vs. 7.8 [6.4–9.2] months, P = 0.001; OS: 10.4 [7.9–15.0] vs. 27.2 [22.3–29.9] months, P < 0.001). Median [95% CI] PFS and OS were also significantly shorter in CTC1C+ than in CTC1C- patients (PFS: 4.3 [3.6–6.0] vs. 8.5 [6.6–10.4], P < 0.001; OS: 7.7 [6.4–13.9] vs. 30.6 [22.6–na], P < 0.001).

Figure 2
figure 2

Progression-free survival and overall survival by CTC status. PFS (left) and OS (right) by CTC status at baseline (top) and after the first cycle of a new line of systemic therapy (bottom) in 356 patients with MBC.

Full size image

Figure 3 shows Kaplan-Meier plots for PFS and OS stratified by change in patients’ CTC status from baseline to completion of the first treatment cycle (CTCKIN). There were significant differences in PFS and OS, depending on CTCKIN (P < 0.001 for PFS and OS). For PFS, we simplified to favorable and unfavorable CTCKIN, depending on CTC1C status. PFS for patients with favorable CTCKIN (i.e. CTCBL- to CTC1C- or CTCBL+ to CTC1C-) did not differ significantly (P = 0.251). Similarly, PFS for unfavorable CTCKIN (i.e. CTCBL- or CTCBL+ to CTC1C+) also showed no significant difference (P = 0.665). Regarding OS, CTCBL status also appeared important since patients with CTCKIN from CTCBL- to CTC1C- lived significantly longer than those with CTCKIN from CTCBL+ to CTC1C- (P = 0.049). OS times for unfavorable CTCKIN did not differ significantly (P = 0.358). When conditioning on non-missing CTC1C values, the median OS time was overestimated by 2.7 months for CTCBL+ and 3.4 months for CTCBL- patients. This provides a rough estimate of the effect of deaths before CTC1C. No CTC1C status was obtained for 12/12, 8/10, and 5/13 patients who died during the first, second, and third month after study entry, respectively. No CTC1C status was obtained for 26/40, 3/5, and 0/6 patients who were censored during the first, second, and third month after study entry, respectively. Table 2 summarizes the results for PFS, OS, and progression by CTCKIN.

Figure 3
figure 3

Progression-free survival and overall survival by CTC KIN . PFS (left) and OS (right) stratified by change in CTC status (CTCKIN) from baseline to completion of the first treatment cycle.

Full size image
Table 2 CTC KIN and association with PFS, OS, and progression at 3-month radiological examination
Full size table

Response and survival

Survival depended significantly on the result of radiological assessment 3 months after inclusion as median [95% confidence interval (CI)] OS times were 29.9 [27.4–37.1] months for patients who achieved at least SD, and 13.6 [9.1–16.4] months for patients with PD (n = 356; P < 0.001).

Multivariate regression analysis

Table 3 shows the result of multivariate regression analysis for PFS and OS using a Cox proportional hazards model including CTCBL, age at study entry, number of metastatic sites, site of metastasis, line of therapy, and molecular subtypes. Significant risk factors for progression were CTCBL+ status, third or higher line of therapy, and TNBC. Significant risk factors for death were CTCBL+, both visceral/local and bone metastases, third or higher line of therapy, and TNBC. The concordance index was 0.62 for the PFS Cox model and 0.71 for the OS Cox model.

Table 3 Cox proportional hazards model with CTC BL
Full size table

Table 4 shows the result of multivariate regression analysis for PFS and OS using a Cox proportional hazards model including CTCKIN. In this model, significant risk factors for both progression and death were CTCBL+ to CTC1C+ kinetics, line of therapy, and TNBC. The presence of both visceral/local and bone metastases was an additional significant risk factor for OS. The concordance index was 0.67 for the PFS and 0.80 for the OS Cox model.

Table 4 Cox proportional hazards model with CTC KIN
Full size table

Discussion

In recent years, several retrospective and a few prospective studies have demonstrated the strong and independent prognostic role of CTCs in MBC [1, 2, 4, 9, 11, 15]. Using the FDA-cleared CellSearch™ system, detection of ≥ 5 CTCs/7.5 ml blood before starting a new line therapy is associated with decreased PFS and OS. In addition, CTCs provide an effective prognostic tool for early response prediction as survival is prolonged once counts ≥ 5 CTCs/7.5 ml blood convert to < 5 CTCs/7.5 ml, i.e. from CTC positive to CTC negative [4, 7, 9, 16]. Thus, serial CTC enumeration promises to provide a fast and easy-to-perform tool for monitoring the efficacy of a given systemic treatment in MBC patients [7]. To address this directly in a clinical setting, the present large study analyzed the changes in CTC status, or CTC kinetics, occurring from baseline to completion of the first cycle of a new line of systemic therapy in patients with MBC. The data were then analyzed to prospectively determine the association of CTC status and first-cycle CTC status with treatment response, PFS, and OS.

Our data demonstrate that patients with favorable CTC kinetics, i.e. those whose CTC status after one cycle of therapy (CTC1C) was negative, were more likely to respond to therapy as determined by RECIST criteria than patients with persistently high CTC counts [5, 8, 16, 17]. Furthermore, PFS was significantly longer in patients with a negative CTC1C status than in those who were CTC positive after completing the first treatment cycle. This observation was independent of the CTC status at baseline, supporting the role of serial CTC enumeration as a means of assessing treatment response. Accordingly, multivariate analysis showed no impact of a positive baseline CTC status on PFS if CTC status turned negative after one cycle of treatment. Budd et al. found CTC assessment to be predictive of survival in both patients with and without radiological progression [5]. They also suggested that CTC assessment might have advantages over radiographic evaluation, including higher reproducibility due to lower interreader variability, useful results at an earlier time, and more robust prediction of survival [5]. Imaging studies, currently the gold standard surrogate for clinical benefit from systemic therapy, are usually not performed before completion of at least two or three cycles of therapy. Hence, CTC determination after one cycle might enable much earlier assessment of treatment response and thus spare patients the unnecessary side effects of ineffective but toxic treatments. Moreover, radiographic imaging is confounded by a considerable degree of intraobserver and interobserver variability, whereas CTC enumeration with the CellSearch™ system is highly standardized [18].

In the current study, the majority of patients (66%) were CTC negative at baseline. This is in contrast to a seminal analysis provided by Cristofanilli et al. [2], who reported 70% of the patients harboring ≥ 5 CTCs/7.5 ml blood. However, in our study, only 31% of patients received third- or higher-line therapy. Thus, the difference might be due to a selection bias.

Other explanations, however, are also conceivable. Despite the prognostic impact of CellSearch CTC in MBC, it has become clear that this technology has limitations. In particular, it is not capable of detecting the entire, highly heterogeneous population of CTCs as it involves EpCAM-based capturing methods [19]. Moreover, a recent retrospective study in 292 MBC patients reported that the probability of undetectable CTCs was increased in patients with negative hormone receptors, high tumor grade, triple-negative disease, and inflammatory breast cancer [20]. The authors suggested that these findings might reflect underestimation of CTCs by CellSearch due partly to CTCs undergoing epithelial-mesenchymal transition (EMT). An earlier study found that a major proportion of CTCs in the blood of MBC patients showed EMT and tumor stem cell characteristics and that such CTCs were associated with an inferior prognosis [21]. On the other hand, it has recently been demonstrated that not all patients with detectable CTCs have a poor prognosis, suggesting that further characterization of these cells might provide more information on their biologic significance. In this regard, Smerage et al. [22] used CellSearch to analyze CTC apoptosis and Bcl-2 expression and show that determination of these markers may have biological and clinical implications. This, therefore, might also offer a further explanation for the large proportion of CTC negative patients in the present study. Moreover, therapeutic regimens might also explain the high CTC negativity rate. A combination of e.g. trastuzumab and lapatinib might be more effective in HER2 positive patients and even stem cell-like cells might be eliminated by such a combination.

In our study, patients with a negative CTC status after the first cycle had a significantly prolonged OS if they were CTC negative at baseline. This observation is in line with results reported by Pierga et al. [9], showing that OS was better in patients with persistently low CTC counts (< 5 CTCs/7.5 ml blood) than in initially CTC positive patients with low CTC counts after one treatment cycle. In addition, it indicates that baseline CTC determination enables identification of more aggressive disease and thus may be valuable in making an early decision whether patients require more aggressive or less aggressive treatment [15]. Of note, the group of baseline positive patients in our study was significantly younger than the baseline negative patients at the time of study entry, although there was no significant difference with respect to age at initial diagnosis. This further supports the hypothesis that higher CTC counts may be suggestive of more aggressive disease in younger women.

Advantages of the CellSearch™ system include semi-automation and proven reproducibility, reliability, sensitivity, linearity, and accuracy [13]. However, it is important to bear in mind that 66% of MBC patients in our cohort had < 5 CTCs/7.5 ml blood at baseline. During the initial phase of the study, which comprised the first 100 patients, CTC status at follow-up was only assessed in patients who had been CTC positive at baseline. Due to the unexpectedly low CTC positivity at baseline, we decided also to evaluate initially CTC negative patients for CTC status at follow-up. However, only 7% of the patients who were CTC negative at baseline were found to be CTC positive after one cycle of treatment. Therefore, it seems that CTC counts, as measured by the CellSearch™ system, are useful as a tool for monitoring treatment efficacy only in patients who are CTC positive when they start a new line therapy, highlighting the need for additional, more sensitive methods of CTC detection. In addition, methods based on the detection of EpCAM, like the CellSearch™ system, might miss CTCs that have undergone epithelial-mesenchymal transition [23].

We found a strong relationship between treatment-associated CTC kinetics and outcome. Favorable CTCKIN was associated with a significantly better disease control rate. In addition, patients with high baseline CTC counts ≥ 5 CTCs/7.5 ml blood that decreased to < 5 CTCs/7.5 ml blood after one cycle of treatment had a PFS similar to patients with baseline counts < 5 CTCs/7.5 ml [24]. In contrast, OS depends not only on the patient’s current CTC status, but also on her previous CTC history. For instance a patient with a CTC1C- status had a better prognosis if she was initially CTCBL- rather than CTCBL+. Thus, a patient’s CTC history might better reflect the overall aggressiveness and prognosis of her breast cancer than the current CTC status alone. Using a somewhat different, CTC count-based approach to classifying CTC kinetics, a recent study by Hartkopf et al. demonstrated that changes in CTC levels from baseline to completion of three treatment cycles also correlated with radiological response and were associated with survival [17]. Median OS was significantly longer in patients with decreasing CTC levels than in patients with increasing CTC counts.

Data from this and other studies [5, 8, 9, 16, 17] do not allow the distinction between breast cancers with unfavorable CTC kinetics that are resistant to the specific type of chemotherapy administered versus those that are resistant to chemotherapy in general. Ongoing prospective trials such as the Southwest Oncology Group (SWOG) protocol S0500 trial and the DETECT III trial will help to shed light on the utility and limitations of measuring CTCs to monitor response to treatment. The SWOG trial randomly assigns MBC patients with persistent CTC counts ≥ 5/7.5 ml blood at the follow-up visit to either continuation of their current therapy or switching to a different regime. DETECT III is a multicenter phase III trial comparing standard therapy +/- lapatinib in HER2 negative MBC patients but with HER2 positive CTCs.

The potential of CTC enumeration and characterization to serve as a “liquid real-time biopsy”, i.e. as a noninvasive means of predicting and monitoring response to treatment in metastatic disease, has recently been comprehensively discussed by Alix-Panabieres and Pantel [25]. Unsuccessful regimens could be abandoned early in favor of alternative regimens, thus sparing patients unnecessary toxicity [6–8]. Moreover, in the future real-time CTC enumeration during therapy should be complemented by additional markers, which enable the monitoring of those cells which possess the highest metastasis-inducing activity within the highly heterogeneous pool of EpCAM+ CTCs [4, 26]. For ER+ luminal MBCs such metastasis-initiating cells have been functionally defined as EpCAM+/CD44+/MET+/CD47+ [4, 26]. However, novel methods have yet to be developed to include these markers in routine clinical practice. Future studies are needed to investigate ways in which CTC enumeration can be combined with computer-assisted assessment of prognosis and adjuvant therapy planning based on various biomarkers [27–29] to further individualize and target the treatment of breast cancer, which remains the most frequent cancer in women in Germany and worldwide [30].

In summary, our study demonstrates that serial CTC monitoring is a versatile tool for predicting treatment outcome in MBC and a useful adjunct to standard diagnostic tests for tailoring therapy. The data presented here further support the hypothesis that the monitoring of CTCs is a promising source of biological information towards predicting the course of disease and its responsiveness to targeted agents, hence paving the way for individualized therapy [24, 25, 31, 32].

Conclusions

CTC status at baseline (CTCBL) and after one cycle of a new line of therapy (CTC1C) and CTC kinetics (CTCKIN, i.e. changes from CTCBL to CTC1C) are highly predictive of outcome in MBC and significantly associated with PFS and OS.

Based on the findings of the present prospective study, we consider serial CTC monitoring a versatile tool for predicting treatment outcome in MBC and a useful adjunct to standard diagnostic tests in tailoring therapy.

Authors’ information

Andreas Trumpp and Andreas Schneeweiss are joint senior authors.